Dickmanns 1991b: The integrated spatiotemporal approach to machine vision, which has allowed outstanding performance with moderate computing power in road vehicle guidance, is extended to general motor vehicle guidance problems on almost flat terrain. Obstacle recognition and relative spatial state estimation by monocular dynamic vision are discussed. A modular vision system architecture centering around image features and physical objects is described. In dynamic scene interpretation conventionally measured variables of the own vehicle like distance traveled or inertial states are exploited in order to disambiguate visual data. Approximate dynamical models for vehicle motion are used in combination with common sense reasoning in order to understand the motion process relative to the environment which is unknown except for some basic properties defined by the task context. Experimental results with road vehicles are given as demonstrations of the general performance capabilities of the method.